Public Trust
Infrastructure

There is a widespread sense that institutions are losing trust. But the diagnosis is usually wrong.

The problem is not that institutions have become more dishonest. It is not that the public has become more cynical, though that is partly true. The deeper problem is structural: the systems through which institutions now act have outpaced the systems through which institutions are held accountable.

For most of the twentieth century, institutional trust was secured through a legible set of mechanisms. Laws defined what was permitted. Professional norms shaped how decisions were made. Audits reviewed what had happened. Elections held officials accountable. These mechanisms were imperfect, but they were coherent. They fit the world they operated in: a world where consequential decisions moved through people, paper, committees, and chains of human authority.

Today, public institutions allocate benefits, enforce rules, distribute money, process applications, triage services, and make determinations through software. Workflows run across vendors, platforms, databases, and third-party systems. Decisions are shaped by algorithms configured years ago, for purposes that may have shifted. And increasingly, AI is being introduced into these workflows: recommending, summarizing, classifying, routing, flagging.

This is not a crisis of intent. Most institutions are not corrupt. The problem is that the infrastructure they operate through cannot show enough of what is happening to sustain trust when it matters: when money is at stake, when a decision affects someone’s livelihood, when a community is asking whether the process was fair.

The traditional accountability model rests on three pillars: documentation, review, and consequence. Record what happened. Review it periodically. Punish violations when found. Each pillar still has a role. None of them, alone or together, is sufficient for modern institutional operations.

Documentation was designed for a world where records were created deliberately: a signed form, a filed report, a meeting minute. Today, the most important records are operational artifacts — log entries, system states, approval timestamps — usually locked inside vendor platforms, overwritten by updates, or simply never structured for review.

Review was designed for a world where the volume of institutional action was manageable. Today, public systems execute thousands of determinations per day. The gap between what happens and what gets reviewed is not a failure of effort. It is a structural condition.

Consequence operates entirely in retrospect. By the time an investigation concludes, the harm has been done. Deterrence is not accountability. It is an acknowledgment that accountability failed.

There is a deeper shift underneath all of this. Modern institutions cannot rely only on trust in authority. They must also support trust in process. Trust in authority says: believe the institution because of who it is. Trust in process says: verify the institution because of what it can show. The first kind of trust is eroding. The second kind has to be built.

Public Trust Infrastructure is the system layer that makes institutional action visible, traceable, reviewable, correctable, and governed. These are not abstractions. Each property has a concrete meaning.

Taken together, these properties define an operating environment in which institutional action can prove itself — not by assertion, but by showing its work.

Every era of institutional modernization has required a new layer. Digital operations required cloud infrastructure. Networked systems required cybersecurity. Regulated industries required compliance frameworks. The AI era requires something none of those layers provides: the capacity to make the operation of consequential systems visible and governable in real time.

Modern institutions need an operating trail. This is different from a paper trail. A paper trail is documentation that accompanies a process. An operating trail is documentation that emerges from a process.

The distinction matters enormously. A paper trail can be constructed after the fact. It can be shaped by what someone wants to preserve or omit. An operating trail is structural — it records events as they occur, in sequence, in context, with the evidence present at each step.

When the story has to be reconstructed, it will always be incomplete, and it will always be reconstructed by someone with a stake in how it reads. That is not accountability. It is a narrative. The two are not the same.

Artificial intelligence is entering institutional workflows in ways that are not yet well understood by the institutions adopting it. AI can summarize cases, recommend decisions, classify applications, flag anomalies, and draft correspondence. These capabilities are real. Some of them are genuinely useful. But they introduce a new class of accountability problem that existing infrastructure is not equipped to handle.

The problem is not that AI makes mistakes. Every system makes mistakes. The problem is that AI can generate outputs without a visible reasoning process, and when those outputs shape institutional decisions, the institution can no longer easily explain what happened or why.

Consider what it means when an AI system recommends that a benefit application be denied, and a staff member approves the recommendation, and the applicant receives a denial letter. What did the AI see? What weight did it assign to different factors? What would have changed the recommendation? If the applicant appeals, what record exists of the basis for the determination?

The relevant question is not whether AI can help. In many cases, it can. The relevant question is: what must the system show before AI is allowed to help? What is the minimum proof layer required to make AI-assisted decisions governable?

The need for public trust infrastructure is visible in almost every domain where institutions act consequentially.

There is a concern, worth addressing directly, that all of this leads to more paperwork, more bureaucracy, more surveillance of workers, and more administrative burden on organizations that are already stretched.

Public trust infrastructure is not that. The goal is not more reporting. The goal is work that can prove itself.

The distinction is between accountability as a parallel activity and accountability as a property of the work itself. In the first model, workers do their jobs and then document what they did — burdensome, often incomplete. In the second model, the system captures what is happening as it happens. The accountability record is not an additional task. It is an output of the operating process.

I grew up in Iran. My husband and I have lived in several countries. We know what it looks like when institutions lose legitimacy — not all at once, but gradually, as the gap between what they claim and what they can show widens beyond the point where public confidence can survive it.

We moved to the United States in part because of the design of its institutions — not their perfection, but their capacity for correction. What is worth protecting here is a set of mechanisms: the ability to question what institutions do, to audit how they do it, to recount contested results, to investigate failures, to challenge decisions through independent processes. That capacity for self-correction is not common. Most of the world does not have it in reliable form. It requires active maintenance.

Modern institutions will not earn trust by saying they are trustworthy. They will earn it by building systems that can show their work.

The technology to do this exists. The conceptual frame exists. What has been missing is the clarity to name the problem precisely and the will to treat it as infrastructure — not as a compliance project, not as a technology upgrade, not as a communications challenge, but as a foundational design requirement for any institution that exercises power over people’s lives with public sanction.